Low levels of hybridization between domestic and wild Mallards wintering in the Lower Mississippi Flyway

Relationships among mtDNA haplotypes were assessed through a median-joining network constructed with Network v. 5.0 (Bandelt et al. 1999). Given that mtDNA haplotypes can be permanently captured through maternal inheritance, we were particularly interested in individuals possessing OW A mtDNA haplotypes as this is a proxy to determine individuals in which their lineage at some point included a female captive-bred Mallard (Lavretsky et al. 2020). This was especially informative for early 2011–12 Mallard samples for which we were unable to assess nuclear variation (see below). Consequently, in addition to visualizing haplotype relationships, we examined OW A versus NW B haplotype ratios across space (i.e., 4 states) and time (2011–12 versus 2020–21).

To evaluate nuclear population structure, we used only the autosomal ddRAD-seq bi-allelic single nucleotide polymorphisms (SNPs). Prior to analyses, we used PLINK v. 0.70 (Purcell et al. 2007) to ensure that singletons (i.e., minimum allele frequency [maf] = 0.0038) and any SNP missing >5% of data across samples were excluded in each dataset. Additionally, we identified independent SNPs by conducting pair-wise linkage disequilibrium (LD) tests across ddRAD-seq autosomal SNPs (--indep-pairwise 2 1 0.5) in which 1 of 2 linked SNPs are randomly excluded if we obtained an LD correlation factor (r2) > 0.5. We conducted all analyses without a priori information on population or species identity.

First, we used the PCA function in PLINK to perform a principal component analysis (PCA). Next, we used ADMIXTURE version 1.3 (Alexander et al. 2009, Alexander and Lange 2011) to attain maximum likelihood estimates of population assignments for each individual, with datasets formatted for the ADMIXTURE analyses using PLINK and following steps outlined in Alexander et al. (2012). We ran each ADMIXTURE analysis with a 10-fold cross-validation, incorporating a quasi-Newton algorithm to accelerate convergence (Zhou et al. 2011). Each analysis used a block relaxation algorithm for point estimation and terminated once the change in the log-likelihood of the point estimations increased by <0.0001. We ran separate ADMIXTURE analyses that included all possible samples and another excluding Khaki Campbell mallards. Each analysis was run for K populations of 1 through 5, and with 100 iterations per each value of K. The optimum K in each analysis was based on the average of cross-validation errors across the iterations per K value; however, we examined additional values of K to test for further structural resolution across analyses. We used the R package PopHelper (Francis 2016) to convert ADMIXTURE outputs into CLUMPP input files at each K value, and to determine the robustness of assignments of individuals to populations at each K value with the program CLUMPP version 1.1 (Jakobsson and Rosenberg 2007). In CLUMPP, we employed the Large Greedy algorithm and 1,000 random permutations. Final admixture proportions for each K value and per sample assignment probabilities (Q estimates; the log-likelihood of group assignment) were based on CLUMPP analyses of all 100 replicates per K value. In addition to the above replicates, standard deviations were calculated under the optimum K population value based on 1,000 bootstraps as implemented in the ADMIXTURE program. Doing so permitted us to evaluate how sensitive our assignment probabilities were given our SNP dataset.

Funding provided by: National Science Foundation
Crossref Funder Registry ID: http://dx.doi.org/10.13039/100000001
Award Number: 2010704